Probing Electron Correlations in Molecules by Two-Dimensional

Feb 21, 2008 - Darius Abramavicius , Jun Jiang , Benjamin M. Bulheller , Jonathan D. Hirst and Shaul Mukamel. Journal of the American Chemical Society...
5 downloads 0 Views 269KB Size
Download PDF
Published on Web 02/21/2008

Probing Electron Correlations in Molecules by Two-Dimensional Coherent Optical Spectroscopy Zhenyu Li,† Darius Abramavicius, and Shaul Mukamel* Department of Chemistry, UniVersity of California, IrVine, California 92697 Received October 1, 2007; E-mail: [email protected]

Abstract: The nonlinear optical signal generated in phenol by three femtosecond pulses with wavevectors k1, k2, and k3 in the phase-matching direction k1 + k2 - k3 is simulated. This two-dimensional coherent spectroscopy (2DCS) signal has a rich pattern containing information on double-excitation states. The signal vanishes for uncorrelated electrons due to interference among quantum pathways and, thus, provides direct signatures of correlated many-electron wavefunctions. This is illustrated by the very different 2DCS signals predicted by two levels of electronic structure calculations: state-averaged complete active space selfconsistent field (SA-CASSCF) and multistate multiconfigurational second-order perturbation theory (MSCASPT2).

1. Introduction

Computing electron correlations is one of the central challenges of electronic structure theory.1-3 At the Hartree-Fock (HF) level, the wave function assumes the form of a single Slater determinant, which is antisymmetric with respect to electron exchange. The probability of finding two electrons with parallel spins at the same point in space thus vanishes (Pauli exclusion). This type of correlation which exists at the HF level is commonly denoted as exchange. The correlation energy of a molecule is defined as the difference between the actual and the HF energy. HF method is the reference point of most quantum chemistry calculations. It is easy to show that the restricted HF method cannot describe the dissociation of molecules into open-shell fragments, e.g., H2 f 2H. This is because weak or broken bonds involve a large occupation of antibonding orbitals, which cannot be described by a single determinant. This type of correlation, which is essential for the correct asymptotic behavior at long distances, is known as static correlation. The remaining, dynamic correlation, is associated with the instantaneous short-range electron-electron interactions. Electron correlation effects are crucial for predicting many molecular properties with high accuracy. For example, uncorrelated HF calculations tend to underestimate bond lengths. Static electron correlation increases the antibonding occupations, leading to too long bond lengths. Only when both static and dynamic correlations are included, we get a closer agreement † Current address: Hefei National Laboratory for Physical Sciences at Microscale, University of Science and Technology of China, Hefei, Anhui 230026, China.

(1) Fulde, P. Electron Correlations in Molecules and Solids, 3rd ed.; SpringerVerlag: Berlin, 1995. (2) Olivucci, M., Ed. Computational Photochemistry; Elsevier: Amsterdam, 2005. (3) Wilson, A. K., Peterson, K. A., Eds. Recent AdVances in Electron Correlation Methodology; American Chemical Society: Washington, DC, 2007. 10.1021/ja0774414 CCC: $40.75 © 2008 American Chemical Society

with experiment. It will thus be highly desirable to develop direct experimental probes for electron correlations. Ideally, such signals should vanish for uncorrelated systems and could thus provide a background-free measure of correlations. A two-dimensional electronic correlation spectroscopy (2DCS) signal, which has precisely this property, has been proposed recently for semiconductor quantum dots.4 Multidimensional techniques have had remarkable success in NMR by probing correlations between coherently excited spins. These techniques provide detailed and specific information on the geometry and dynamics of complex molecules. In recent years, these ideas have been extended to femtosecond laser spectroscopy.5 Infrared techniques probe correlations among the amide vibrations in proteins,6 and visible pulses probe exciton dynamics in aggregates such as photosynthetic complexes7,8 and in semiconductor quantum wells.9 The proposed technique uses four femtosecond pulses with wavevectors k1, k2, k3, and k4 ) k1 + k2 - k3 and detects the heterodyne signal along k4. This technique which involves double-excitation states has been proven useful for electronic and vibrational excitons.10-15 Most recently, we have demonstrated that this pulse sequence can be used to directly probe electron correlations.4 Since the signal vanishes at the uncorrelated level, it provides a novel, unusually sensitive, probe for electron correlations. (4) Mukamel, S.; Oszwaldowski, R.; Yang, L. J. Chem. Phys. 2007, 127, 221105. (5) Mukamel, S. Ann. ReV. Phys. Chem. 2000, 51, 691. (6) Zanni, M. T.; Hochstrasser, R. M. Curr. Opin. Struct. Biol. 2001, 11, 516. (7) Lee, H.; Cheng, Y.-C.; Fleming, G. R. Science 2007, 316, 1462. (8) Engel, G. S.; Calhoun, t. R.; Read, E. L.; Ahn, T.-K.; Mancal, T.; Cheng, Y. C.; Blankenship, R. E.; Fleming, G. R. Nature 2007, 446, 782. (9) Li, X.; Zhang, T.; Borca, C. N.; Cundiff, S. T. Phys. ReV. Lett. 2006, 96, 57406. (10) Leegwater, J. A.; Mukamel, S. J. Chem. Phys. 1994, 101, 7388. (11) Kuhn, O.; Chernyak, V.; Mukamel, S. J. Chem. Phys. 1996, 105, 8586. (12) Scheurer, C.; Mukamel, S. J. Chem. Phys. 2001, 115, 4989. (13) Mukamel, S.; Tortschanoff, A. Chem. Phys. Lett. 2002, 357, 327. (14) Cervetto, V.; Helbing, J.; Bredenbeck, J.; Hamm, P. J. Chem. Phys. 2004, 121, 5935. (15) Fulmer, E. C.; Mukherjee, P.; Krummel, A. T.; Zanni, M. T. J. Chem. Phys. 2004, 120, 8067. J. AM. CHEM. SOC. 2008, 130, 3509-3515

9

3509

Li et al.

ARTICLES

In this paper we demonstrate the power of this technique by high-level electronic structure calculations of a simple molecule, phenol. Quantum chemistry techniques for ground state properties are well developed. Excited-state properties are more difficult to calculate. We shall use methods which can treat the ground and excited states in a balanced way. There are two types of approaches for the electronic structure of excited states. The first is the propagator approach (also known as Green’s function, equation-of-motion, or linear response theory), which treats the excited states by perturbing the ground electronic state by the electromagnetic field. These techniques strongly depend on the quality of ground state calculation. Time-dependent density functional theory (TDDFT) is a popular example of such methods. The other class of approaches are wave function based. Starting with a HF ground state wavefunction given by a Slater determinant of molecular orbitals (MOs), one can construct excited states by promoting one or more electrons from an occupied to an unoccupied orbital. With the independent particle Hamiltonian (IPH), the excited-state energies are given by differences of orbital energies. In this uncorrelated HF-IPH picture, the transition energy of a double-excitation state is simply given by the sum of the corresponding two singleexcitation energies. A more rigorous description is configuration interaction singles (CIS),16 where the excited states are expressed as linear combinations of all singly excited determinants formed by replacing an occupied by an unoccupied orbital. Here, the ground state is still described by the HF Slater determinant, and all excitations are based on this reference configuration. Considering only singly excited determinants may be a serious approximation. Linear combination of all possible excited determinants (full configuration interaction, FCI) gives the most accurate wavefunction for a specific atomic basis set. However, such calculations are only feasible for very small systems, since the time and memory requirements grow exponentially with size. For large molecules, they can be carried out only by restricting the calculation to a small active space. The molecular orbitals should then be optimized together with the CI coefficients. This type of multiconfigurational self-consistent field (MCSCF) method is denoted a complete active space self-consistent field (CASSCF).17 To obtain good orbitals for both ground state and excited states, CASSCF is usually performed by minimizing the average energy of a few states of interest. This is known as state averaged (SA) CASSCF. CASSCF describes the static correlations very well and usually generates an optimal set of MOs together with reasonable ground and excited state wavefunctions. Dynamic correlations can be further included by multireference configuration interaction (MRCI) or a multiconfigurational perturbation theory. In MRCI, the CI expansion comprises all configurations that can be generated by single and double excitations from the CASSCF reference configurations. MRCI is only feasible in relatively small molecules such as benzene. Furthermore, it is not size-extensive. Dynamic correlations can be alternatively included perturbatively. Moller-Plesset second-order perturbation theory (MP2) for ground states with a single determinant (16) Foresman, J. B.; Head-Gordon, M.; Pople, J. A.; Frisch, M. J. J. Phys. Chem. 1992, 96, 135. (17) Roos, B. O.; Taylor, P. R. Chem. Phys. 1980, 48, 157. 3510 J. AM. CHEM. SOC.

9

VOL. 130, NO. 11, 2008

Figure 1. Active orbitals 1a′′ to 4a′′ (top row) and 5a′′ to 7a′′ (second row) used in our simulations of phenol. 4a′′ is the HOMO, and 5a′′ is the LUMO. The last panel shows the optimized geometry.

reference function has been extended to multiple reference functions of the CAS type, and it is called the CASPT2 method.18-20 For systems with a strong interaction between different CASSCF wave functions, an effective Hamiltonian can be constructed for several selected states, and the multistate (MS) CASPT2 energies can be then obtained by diagonalizing this Hamiltonian.21 We shall employ transition energies and dipoles obtained by two levels of theory, SA-CASSCF and the higher-level MSCASPT2, to simulate the electronic absorption and 2DCS spectra. The 2DCS signals show projections of the correlated doubly excited wavefunctions onto the manifold of singly excited states. They are very different for the two levels of theory, reflecting the different levels where correlations are treated by the two techniques. The remainder of this article is organized as follows. In section 2, we introduce the computational methods. The excited states and the linear absorption of phenol are presented in section 3. The 2DCS technique is described in section 4, and the simulated signals are shown in section 5. Finally, we conclude in section 6. 2. Computational Methods The geometry of phenol, optimized at the B3LYP/6-311++G** level of density functional theory (DFT) using the Gaussian 03 package,22 is shown in Figure 1. The ground state and excited states electronic structures were then calculated at the SA-CASSCF/MS-CASPT2 levels using the MOLCAS 7.0 package23 and an atomic natural orbital (ANO) basis set.24 For heavy atoms, the primitives were contracted to triple-ζ plus polarization quality (4s3p1d), and for hydrogen atoms, the contraction was of double-ζ plus polarization (2s1p) quality. This basis set was further augmented with a set of off-atom diffuse functions for Rydberg states. Following Roos et al.,25 the following steps were (18) Andersson, K.; Malmqvist, P.-A.; Roos, B. O. J. Chem. Phys. 1992, 96, 1218. (19) Andersson, K.; Roos, B. O. Chem. Phys. Lett. 1992, 191, 507. (20) Roos, B. O.; Andersson, K.; Fulscher, M. P. J. Chem. Phys. 1992, 192, 5. (21) Finley, J.; Malmqvist, P.-A.; Roos, B. O.; Serrano-Andres, L. Chem. Phys. Lett. 1998, 288, 299. (22) Frisch, M. J. et al. Gaussian 03, revision C.02; Gaussian, Inc.: Wallingford, CT, 2004. (23) Karlstrom, G.; Lindh, R.; Malmqvist, P.-A.; Roos, B. O.; Ryde, U.; Veryazov, V.; Widmark, P.-O.; Cossi, M.; Schimmelpfennig, B.; Neogrady, P.; Seijo, L. Comput. Mater. Sci. 2003, 28, 222. (24) Widmark, P.-O.; Malmqvist, P.-A.; Roos, B. O. Theor. Chim. Acta 1990, 77, 291. (25) Roos, B. O.; Fulscher, M. P.; Malmqvist, P.-A.; Merchan, M.; SerranoANdres, L. In Quantum Mechanical Electronic Structure Calculations with Chemical Accuracy; Kluwer Academic Publisers: Dordrecht, 1995; pp 357-438.

Probing Electron Correlations in Molecules

ARTICLES

Table 1. Excitation Energies (in eV) of the First Six Excited States of Phenola

2 A′ 3 A′ 4 A′ 5 A′ 6 A′ 7 A′

Platt notation

benzene notation

main configuration

SA-CASSCF

MS-CASPT2

7S-CASPT2

expt36

Lb La Bb Ba

B2u B1u E1u E1u E2g E2g

HfL H f L+1 H-1 f L+1 H-1 f L H-2 f L H-2 f L+1

4.81 7.35 8.95 8.79 7.85 8.08

4.89 6.16 6.96 7.05 8.00 8.19

4.79 5.93 6.66 6.80 8.10 8.04

4.59 5.82 6.70 6.93

a The two columns denoted by SA-CASPSCF and MS-CASPT2 are results from calculations of the first 20 states. 7S-CASPT2 denotes a MS-CASPT2 calculation for the lowest seven states, where a 0.1 hartree level shift was used. States are shown in ascending order of the MS-CASPT2 energies.

performed in order to obtain these Rydberg basis functions. First, a CASSCF calculation was performed for the phenol cation, and the charge center was set to the center of the Rydberg basis functions. This calculation was then repeated with a (8s8p8d) basis set for Rydberg states26 added. Finally, Rydberg orbitals were generated using the GENANO program in the MOLCAS package. The orbital coefficients were used as contraction coefficients of the final (1s1p1d) Rydberg basis set. The quality of the CASSCF calculation depends crucially on the chosen active space. Since the low-energy excitations of phenol are π to π* transitions, we included all π orbitals and electrons in our active space. With Cs symmetry, this leads to eight electrons in seven a′′ orbitals. However, for small molecules in the gas phase, Rydberg and valence excited-state energies are very close. Rydberg states should thus be also included in the active space. An earlier study had indicated that the Rydberg-valence mixing in phenol is weak.27 The following strategy was therefore used in order to exclude the Rydberg orbitals from the active space. First, the sp Rydberg orbitals were included in the active space. A CASSCF calculation was performed, and the optimized Rydberg states were then deleted. The d Rydberg states were optimized and deleted in the same way. The final seven active orbitals which have no Rydberg character are shown in Figure 1. Within this active space, the lowest 20 A′ singlet states were calculated at the SA-CASSCF level, where the CI and MO coefficients are optimized to minimize the average energy of the 20 states. An MSCASPT2 calculation was then performed to include dynamical correlations. The modified zeroth-order Hamiltonian28 with a IPEA shift of 0.25 hartree was used. A 0.3 hartree level-shift was also employed to eliminate the intruder states. The transition moments between the 20 states are calculated by the CAS state interaction (CASSI) method29 implemented in the RASSI program in the MOLCAS package. At the MS-CASPT2 level, perturbatively modified CASSCF wave functions and MS-CASPT2 energies are used for the CASSI calculation. Using these transition energies and dipoles, the spectra were simulated by the sum over states (SOS) response function expressions30,31 implemented in the SPECTRON package.32 The line shape is obtained by cumulant expansion of the Gaussian fluctuation (CGF) formula using an overdamped Brownian oscillator spectral density.30 We used a bath temperature of 300 K and time scale of 1 ps. The coupling parameter was set to give a Gaussian profile with a σ ) 200 cm-1 variance. An average over all possible orientations of the molecule was performed during the spectra simulations.

3. Excited States and the Linear Absorption of Phenol

The excited states of phenol have been extensively studied, both theoretically27,33,34 and experimentally.35-38 The first two singlet excited states mainly correspond to transitions from the highest occupied molecular orbital HOMO (H) to the lowest unoccupied molecular orbital LUMO (L), and L+1. Their corresponding Platt’s notation39 is Lb and La, where the subscripts refer to the transition dipole orientation, a for the long axis and b for the short axis. These two states are the analogues of the B2u and B1u states of benzene. The leading

configuration of the next two excited states Bb and Ba are H-1 to L+1 and to L. These are the analogues of the E1u states of benzene. Some properties of the six lowest excited states are listed in Table 1. The first three excited-state energies calculated by MSCASPT2 are ∼0.3 eV higher than the experimental absorption peaks. We note that the vertical excitation energies for the first two excited states of benzene are 0.1 eV higher than the absorption band maxima.40 Assuming a similar blue shift between vertical excitation energy and absorption band maxima for phenol, the actual difference between theory and experiment is ∼0.2 eV. We aim toward a balanced description for the low single-excitation and higher double-excitation states. For higher excited states, a much larger active space will be required. Therefore, an MS-CASPT2 calculation only for low-energy excited states will give better agreement with experiment. Our test MS-CASPT2 calculation with the current active space but only for the lowest seven states (7S-CASPT2 in Table 1) gives good agreement with experiment. The higher excited states are listed in Table 2. Most of them are dominated by double-excitation configurations. The SACASSCF states more strongly mix different configurations. Unlike the lowest six excited states, there is no one-to-one map between SA-CASSCF and MS-CASPT2 for the higher states based on their main configurations. This implies that doubleexcitation states are more sensitive to electron correlations and their calculation is more challenging. A good indicator of electron correlations is the anharmonicity parameter, defined as the difference of the transition energy of a double-excitation state and the sum of the two corresponding single-excitation states. The anharmonicity vanishes for uncorrelated electrons, where excitation energies are additive. For (26) Kaufmann, K.; Baumeister, W.; Jungen. M. J. Phys. B: At. Mol. Opt. Phys. 1989, 22, 2223. (27) Rogers, D. M.; Hirst, J. D. J. Phys. Chem. A 2003, 107, 11191. (28) Ghigo, G.; Roos, B. O.; Malmqvist, P. A. Chem. Phys. Lett. 2004, 396, 142. (29) Malmqvist, P.-A.; Roos, B. O. Chem. Phys. Lett. 1989, 155, 189. (30) Mukamel, S. Principles of Nonlinear Optical Spectroscopy; Oxford University Press: New York, 1995. (31) Mukamel, S.; Abramavicius, D. Chem. ReV. 2004, 104, 2073. (32) Zhuang, W.; Abramavicius, D.; Hayashi, T.; Mukamel, S. J. Phys. Chem. B 2006, 110, 3362. (33) Lorentzon, J.; Malmqvist, P.-A.; Fulscher, M.; Roos, B. O. Theor. Chim. Acta 1995, 91, 91. (34) Zhang, L.; Peslherbe, G. H.; Muchall, H. Photochem. Photobiol. 2006, 82, 324. (35) Martinez, S. J., III; Alfano, J. C.; Levy, D. H. J. Mol. Spectrosc. 1992, 152, 80. (36) Kimura, L.; Nagakura, S. Mol. Phys. 1965, 9, 117. (37) Ratzer, C.; Kupper, J.; Spangenberg, D.; Schmitt, M. Chem. Phys. 2002, 283, 153. (38) Dearden, J. C.; Forbes, W. F. Can. J. Chem. 1956, 37, 1294. (39) Platt, J. R. J. Chem. Phys. 1949, 17, 484. (40) Bernhardsson, A.; Forsberg, N.; Malmqvist, P. A.; Roos, B. O.; SerranoAndres, L. J. Chem. Phys. 2000, 112, 2798. J. AM. CHEM. SOC.

9

VOL. 130, NO. 11, 2008 3511

Li et al.

ARTICLES Table 2. Higher Excited Statesa SA-CASSCF

Ex (eV)

8 A′ 9 A′ 10 A′ 11 A′ 12 A′ 13 A′ 14 A′ 15 A′ 16 A′ 17 A′ 18 A′ 19 A′ 20 A′ a

Ea (eV)

11.13 11.43 11.59 11.80 11.82 12.82 13.22 13.32 13.62 14.66 14.72 15.03 16.09

-0.57

-4.36 -2.68

MS-CASPT2 main configurations

Ex (eV)

H-1,H f L,L+1; 2Hf2L 2H f 2L; 2H-1 f 2L; 2H f 2L+1 2H f L,L+1 H-2 f L+1 H-2 f L; H-1 f L+3; 2H f L,L+1 H-3 f L 2H-1 f 2L H-3 f L+1 H-1,H f 2L+1 2H-1 f 2L+1; 2H f 2L H-1,H f 2L; H-1,H f 2L+1 H-2 f L+2 H f L+2; H-1,H f L,L+1

9.39 9.87 10.18 10.43 10.63 10.87 11.23 11.67 11.94 12.32 12.55 12.82 13.34

Ea (eV)

main configurations

H f L+2 H-3 f L H-1,H f L,L+1 2H f L,L+1 H-3 f L 2H f 2L; 2H f 2L+1 H-3 f L+1 2H-1 f 2L H-1,H f 2L+1 H-1,H f 2L H-1,H f 2L; 2H f 2L+1 H-2 f L+2 2H-1 f 2L+1

-0.62

-2.43 -1.18 0.38 -0.58

Ex is the transition energy, and Ea is the anharmonicity. Table 3. Low-Energy Linear Absorption Properties of Phenola

HF-IPH TDDFT SA-CASSCF MS-CASPT2 expt36 a

Lb

La

La − Lb

r

12.05 5.05 4.81 4.89

12.62 5.81 7.35 6.16

0.57 0.76 2.54 1.27

1.17 0.85 1.81 1.73

4.59

5.82

1.23

6.6

r is the intensity ratio of the La and Lb transitions. All energies are in

eV.

Figure 2. Simulated absorption spectrum for phenol in the gas phase. Energy ranges 4.6 to 7.6 eV in the SA-CASSCF spectrum and 4.5 to 6.5 eV in the MS-CASPT2 spectrum. Shadowed areas mark the spectral region covered by the pulse bandwidths.

phenol, the 2H f L, L+1 state, which is a combined state of the first two single-excitation states (Lb and La), can be clearly identified by both SA-CASSCF (10A′) and MS-CASPT2 (11A′). The predicted anharmonicities are -0.57 and -0.62 eV, respectively. The absorption spectra are displayed in Figure 2. In addition to SA-CASSCF and MS-CASPT2, we also show two additional levels of theory. One is TDDFT (B3LYP/6-311G**), where the diffusion functions were excluded from the basis set to avoid valence Rydberg mixing. Within the adiabatic approximation, it contains correlations but not double-excitation states.41,42 The other is the uncorrelated HF-IPH model, where the ground state wavefunction was calculated at the HF/6-311G** level. The (41) Maitra, N. T.; Zhang, F.; Cave, R. J.; Burke, K. J. Chem. Phys. 2004, 120, 5932. (42) Casida, M. J. Chem. Phys. 2005, 122, 54111. 3512 J. AM. CHEM. SOC.

9

VOL. 130, NO. 11, 2008

HF-IPH transition dipoles were obtained by calculating the dipole integral between two single determinants. The lowest two peaks in all spectra correspond to the Lb and La states. As shown in Table 3, the TDDFT and MS-CASPT2 calculations give relatively good excitation energies compared to experiment, and HF-IPH strongly overestimate the excitation energies. TDDFT predicts a stronger Lb peak, in contrast to experiment and to SA-CASSCF and MS-CASPT2. Both HF-IPH and TDDFT underestimate the energy splitting between Lb and La. SA-CASSCF overestimates it, and MS-CASPT2 is in a very good agreement with experiment. The absorption spectra provide some valuable information about correlations through the peak positions and intensities. In the next section, we shall demonstrate how 2DCS provide a much more detailed look into the correlated many-body wavefunctions. 4. 2DCS Signal Induced by Electron Correlations

The proposed experiment shown in Figure 3 uses four laser pulses with wavevectors k1, k2, k3, and k4 ) k1 + k2 - k3. The signal is defined as the change in the k4 beam transmitted intensity due to the interference with the induced nonlinear polarization. This configuration, known as heterodyne detection, can be viewed as a stimulated four-wave mixing signal.43 We assume an impulsive experiment with temporally well-separated pulses. The molecule-field interaction is -Vˆ E(r, t), where Vˆ is the dipole operator, and the field is expanded as

E(r, t) ) 4

∑j j(t - τj)eik r-iω (t-τ ) + /j (t - τj)e-ik r+iω (t-τ ) j

j

j

j

j

j

(1)

where the jth pulse is centered at time τj and has an envelope (43) Marx, C.; Harbola, U.; Mukamel, S. Phys. ReV. A 2008, in press.

Probing Electron Correlations in Molecules

ARTICLES

Figure 3. Top panel: Pulse configuration for 2DCS. k1, k2, and k3 are the input pulses. The signal is generated along the detection pulse k4. Bottom panel: The pulse sequence. tj are the time intervals between pulses centered at τi.

j(t - τj), carrier frequency ωj, and wavevector kj. The consecutive time delays between the pulses in chronological order are denoted t1, t2, and t3. The signal is given by

Figure 4. (a) Many-body states of phenol, including the ground state |g〉 (black), the manifold of single-excitation states |e〉 (red), and the manifold of double-excitation states |f〉 (blue). States off-resonant with fields are marked by dash lines. The shadow areas mark the energy range covered by pulse bandwidths for single- and double-photon absorption. (b) Schematic level model for an uncorrelated (harmonic) system.

I(3)(τ4, τ3, τ2, τ1) ) -Im

∫-∞∞ dτj4 ∫-∞τj dτj3 ∫-∞τj dτj2 ∫-∞τj dτj1R(3)(τj4, τj3, τj2, τj1) 4

3

2

× /4(τj4 - τ4)/3(τj3 - τ3)2(τj2 - τ2)1(τj1 - τ1) × e+iω4τj4+iω3τh3-iω2τj2-iω1τj1 (2) where the response function R(3) is given by a combination of the four-time correlation functions of the dipole operator Vˆ (τ), which can be calculated by the sum-over-states expression.30 In principle, the entire complex manifold of excited states must be included in computing R(3). However, in practice, only some groups of states are relevant for resonant signals. These states can be controlled by selecting the phase-matching direction (k4), the carrier frequencies, and the pulse bandwidths, as will be explained below. For simplicity, in our simulations, we assumed the same carrier frequencies for all pulses ω1 ) ω2 ) ω3 ) ω4 ≡ ω0 and used rectangular spectral envelopes for all pulses. ω0 was fixed in the middle of the two peaks in the linear absorption. For MS-CASPT2 excited states, this gives four pulses centered at 5.5 eV with a bandwidth ∆ ) 2 eV. For SA-CASPT2, the four pulses are centered at 6.1 eV with ∆ ) 3 eV. With this pulse configuration, as shown in Figure 4a, the first two MSCASPT2 excited states of phenol, Lb and La, correspond to one photon energy. They are single-excitation states, denoted by |e〉. The signal generated in the k1 + k2 - k3 direction has two contributions, represented by the double-sided Feynman diagrams30 A and B, shown in Figure 5. These diagrams represent the resonant contributions which dominate the signal. In both diagrams, the molecule first absorbs an ω1 photon, which brings it to state |e〉. The density matrix element |e〉 〈g| oscillates at frequency ωeg. The accessible |e〉 state must lie in the region ω0 ( ∆/2 allowed by the pulse bandwidth. Pulse k2 is then absorbed and brings |e〉 to one of the states |f〉 in the region 2ω0 ( ∆. The density matrix element |f〉 〈g| oscillates at frequency ωfg. So far, the sequence of events is the same for both diagrams A and B. The two differ however by the action of the third pulse. In A, it stimulates an emission bringing the molecule to a state |e′〉 in the region ω0 ( ∆/2. In B, a photon

Figure 5. (A and B) The two double sided Feynman diagrams contributing to the k1 + k2 - k3 2DCS signal. (C) Diagram representing the k1 + k2 + k3 2DCS signal. Both signals are induced by electron correlations.

is absorbed bringing the density matrix to an |f〉 〈e′| coherence which oscillates at the ωfe′ frequency. There are two ways to define double-excitation states. Using the electronic structure, these are excited states whose dominant configuration has two electrons promoted from occupied to unoccupied molecular orbitals. Spectroscopically, any state located in the two-photon energy region ω1 + ω2 ( ∆ is denoted a double-excitation state. Its contribution to the spectrum is determined by its transition dipole from the lower singleexcitation states. For uncorrelated electrons with zero anharmonicity, these two definitions coincide. As shown in Figure 4b, in this case, the double-excitation states are combinations of two single-excitation states, and the corresponding excitation energy is the sum of the two single-excitation energies. We define the multidimensional signal by the triple Fourier transform

S(3)(Ω3, Ω2, Ω1) )

∫ ∫ ∫0∞ dt3 dt2 dt1 eiΩ t +iΩ t +iΩ t 11

22

33

×

I (t1 + t2 + t3, t1 + t2, t1, 0) (3) (3)

For well-separated pulses, the signal is given by44 J. AM. CHEM. SOC.

9

VOL. 130, NO. 11, 2008 3513

Li et al.

ARTICLES S(3)(Ω3, Ω2, Ω1) )

∑

1(ω1 - ωeg)2(ω2 - ωfe)/3(ωfe′ - ω3)/4(ωe′g - ω4)Vge′Ve′fVfeVeg

e,e′,f

∑ e,e′,f

-

(Ω3 - ωe′g + iΓe′g)(Ω2 - ωfg + iΓfg)(Ω1 - ωeg + iΓeg) 1(ω1 - ωeg)2(ω2 - ωfe)/3(ωe′g - ω3)/4(ωfe′ - ω4)Ve′fVfeVegVge′ (Ω3 - ωfe′ + iΓfe′)(Ω2 - ωfg + iΓfg)(Ω1 - ωeg + iΓeg) (4)

where j(Ωj) ) ∫ dt j(t) eiΩjt, ωµν ≡ Eµ - Eν is the frequency, Γµν is the dephasing rate of the transition between states ν and µ, and ω4 ) ω1 + ω2 - ω3. The two terms correspond, respectively, to diagrams A and B. By holding ω1, ω2, and ω3 fixed (this selects the relevant energy range) and calculating the signal versus Ω1, Ω2, and Ω3, we obtain a three-dimensional signal. We shall present the signal in two types of two-dimensional plots. The first is (Ω1, Ω2) for a fixed t3. Note that this signal vanishes for t3 ) 0. The second is (Ω3, Ω2) for a fixed t1. By varying the time delay t3 for the (Ω1, Ω2) signal or t1 for the (Ω3, Ω2) signal, we can extract information about electron dynamics. As shown in Figure 4b, for uncorrelated electrons, the transition energy of a double-excitation state will be equal to the sum of the corresponding two single-excitation states. For a double-excitation state |f〉 ) |e1e1〉, e ) e′ ) e1 in both diagram A and B. We have ωe′g ) ωfe′ ) ωge1, and the two diagrams exactly cancel, and thus S(3) ) 0. For the excited state |f〉 ) |e1e2〉, the cancellation of A and B is more subtle. Setting e ) e1, then e′ will be either e1 or e2 for a harmonic system without correlation. If we chose e′ ) e1 in diagram A, it will be canceled by diagram B with e′ ) e2. This can be easily shown by noting that ωe1g ) ωfe2. Similarly, if e′ ) e2 in diagram A and e′ ) e1 in diagram B, they also cancel. Therefore, the entire 2DCS signal is induced by electron correlations and Vanishes for uncorrelated systems. 5. 2DCS of Phenol

The simulated (Ω1, Ω2) signal for t3 ) 1 fs is shown in Figure 6. SA-CASSCF and MS-CASPT2 show a rich and clearly distinct peak pattern. The contributions from diagrams A and B are similar for each quantum chemistry level. Generally, the MS-CASPT2 signal, which includes higher level electron correlations, shows more peaks. The Ω1 axis represents the single-excitation states. Two single-excitation states Lb and La are accessible by our pulse bandwidths, yielding a two-stripe pattern for the 2DCS signal. The Ω2 axis represents the doubleexcitation states. These are clearly seen. Generally, transitions from La to double-excitation states are stronger than transitions from Lb. The strongest MS-CASPT2 peak corresponds to the transition from La to the 2H-1 f 2L state (15A′). The strongest SA-CASSCF peak is from the same transition (La to 14A′). It is interesting to note that the strongest peaks in our 2D signals do not come from the combination state of Lb and La (2H f L, L+1). This indicates that electron correlations in phenol are very strong and the level scheme is highly anharmonic. A notable advantage of 2DCS is its ability to provide direct information on the many-body correlated waVefunction of double-excitation states, by looking at the relative strength of peaks representing transitions from different single-excitation (44) Schweigert, I. V.; Mukamel, S. Unpublished. 3514 J. AM. CHEM. SOC.

9

VOL. 130, NO. 11, 2008

Figure 6. Simulated S(3)(t3, Ω2, Ω1) 2DCS signal (amplitude) of phenol at t3 ) 1 fs. Left column: SA-CASSCF. Right column: MS-CASPT2. From top to bottom, contribution from diagrams A,B and the total signal. The strongest peak in the total-signal panel is indicated by a blue circle. The MS-CASPT2 double-excitation states, 17A′, is marked by a dashed line.

states. Let us consider the MS-CASPT2 double-excitation state at 12.32 eV (17A′). It has a much stronger transition from Lb than from La. This can be easily understood by looking at its configuration coefficients. The main configuration of 17 A′ is H-1, H f 2L, which is a combination of H-1 f L and H f L (Lb). For systems with weak anharmonicity, such as photosynthetic complexes described by the Frenkel exciton model, it is possible to directly extract the main configuration of the double-excitation states by this type of analysis.45 The (Ω3, Ω2) signal displayed in Figure 7 is more complex. The contribution of diagram A is similar to that in Figure 6. However, diagram B leads to a more complicated signal. Since the Ω3 axis now represents transition energies between singleand double-excitation states ωfe′, the signal does not show a simple two-stripe pattern. Even though both 2D signals are induced by correlation, they carry a different information. The S(3)(t3, Ω2, Ω1) signal is simpler and directly reveals the energies of the relevant singleand double-excitation states. With this information, combined with wavefunction projection analysis from both S(3)(Ω3, Ω2, t1) and S(3)(t3, Ω2, Ω1) signals, we obtain additional information about the excited state wavefunction. 6. Conclusions and Discussion

A novel two-dimensional coherent spectroscopy technique has been demonstrated to provide a direct probe of electron correlations in phenol. The signals obtained using excited states calculated at the SA-CASSCF and MS-CASPT2 levels are (45) Abramavicius, D.; Mukamel, S.; Voronine, D. V. Unpublished.

Probing Electron Correlations in Molecules

ARTICLES

by scanning the carrier frequencies over the entire frequency range. In fact, any three states |e〉, |e′〉, and |f〉 can be probed by tuning the carrier frequencies of the four pulses. By combining spectra obtained with various carrier frequencies, it should be possible to reconstruct the signal simulated here. When the pulses are tuned to ω4 ≈ ωe′g, ω3 ≈ ωfe′, ω2 ≈ ωef, and ω1 ≈ ωge, we obtain the (ωeg, ωfg, ωe′g) peaks corresponding to diagram A. The peaks (ωeg, ωfg, ωfe′) from diagram B will be probed with ω4 ≈ ωfe′, ω3 ≈ ωe′g, ω2 ≈ ωef, and ω1 ≈ ωge. Attosecond X-ray pulses have much broader bandwidths (e.g., 1 eV for 1 fs pulse).46,47 X-ray Raman techniques44 could thus probe the entire 2D signal in a single shot. It should be noted that the cancellation predicted here is purely electronic. Coupling to other degrees of freedom (vibrations and dephasing induced by solvent) can affect different excited states differently and may break the exact cancellation. This could result in weak new features. Finally, we note that, in addition to the k1 + k2 - k3 signal discussed here, the 2DCS signal in the k1 + k2 + k3 direction also vanishes for uncorrelated electron systems. The sum-overstates expression of the signal is (see diagram C in Figure 5) S(3) h (Ω3, Ω2, Ω1) )

∑ e,f,h

1(ω1 - ωeg)2(ω2 - ωfe)3(ω3 - ωhf)/4(ωhg - ω4)VghVhfVfeVeg (Ω3 - ωhg + iΓhg)(Ω2 - ωfg + iΓfg)(Ω1 - ωeg + iΓeg) (5)

Figure 7. Simulated S(3)(Ω3, Ω2, t1) signal (amplitude) of phenol at t1 ) 0. Left panel: SA-CASSCF. Right panel: MS-CASPT2. Contributions from diagrams A,B and the total signal are shown, from top to bottom. The same color bar as that in Figure 6 is used.

compared. The former only includes static correlations, whereas the latter contains dynamic correlations as well. The 2DCS signals predicted by these two levels of theory are very different. This signal provides information about transition energies of double-excitation states, as well as their many-body correlated wavefunctions. The ability to measure double-excitation states directly can provide a new experimental test for the accuracy of the electron correlations described in different levels of theory and offers a way for visualizing some projections of the manyelectron wavefunctions. Since an excited configuration is a molecular-orbital-based concept, we do not expect to unambiguously determine the excited configurations experimentally, especially for molecules with a strong mixing between singly and doubly excited configurations. However, we can obtain some useful information about the excited-state wavefunction by analyzing the projections of the two-excitation state onto the single-excitation states. To simplify the simulations and generate the entire spectrum in a single calculation, we have used unphysically broad pulse bandwidth (2 eV for MS-CASPT2 and 3 eV for SA-CASSCF) in the 2DCS signal simulation. In practice, the predicted signal can be observed piecewise by multiple measurements obtained

where the summation runs over single-excitation states |e〉, double-excitation states |f〉, and triple-excitation states |h〉. According to the Condon-Slater rules, the transition dipole between two Slater determinants which differ by two or more orbitals is zero. Therefore, for uncorrelated electrons, Vgh ) 0. Three excitations bring the molecule to a three quantum coherence, and a dipole moment Vgh is required to bring it back to a population and generate a signal. The signal S(3) h thus vanishes for uncorrelated electrons. In this case however, there is only one pathway; the vanishing of the signal is not caused by the destructive interference between two pathways. S(3) h could probe electron correlations corresponding to both double excitations and triple excitations. Acknowledgment. We wish to thank Prof. B. O. Roos for help in using the MOLCAS package. This work was supported by the National Institutes of Health Grant GM59230 and the National Science Foundation Grant CHE-0446555. Supporting Information Available: The optimized geometry and total energy of phenol molecule, the expressions for the 2D signals, and the complete citation of ref 22. This material is available free of charge via the Internat at http://pubs.acs.org. JA0774414 (46) Goulielmakis, E.; Yakovlev, V. S.; Cavalieri, A. L.; Uiberacker, M.; Pervak, V.; Apolonski, A.; Kienberger, R.; Kleineberg, U.; Krausz, F. Science 2007, 317, 769. (47) Kapteyn, H.; Cohen, O.; Christov, I.; Murnane, M. Science 2007, 317, 775.

J. AM. CHEM. SOC.

9

VOL. 130, NO. 11, 2008 3515